Method and system for bridge detection by using unmanned aerial vehicle group

Abstract

The embodiment of the invention provides a method and system for bridge detection by using an unmanned aerial vehicle group, and belongs to the technical field of bridge structure detection. The method comprises the following steps: forming a seam segment matrix for measuring task generation based on initialized acquisition data; generating a multi-time scale measurement task set based on each target time scale; performing unmanned aerial vehicle distribution according to the multi-time scale measurement task set, determining a flight window of each unmanned aerial vehicle under each time scale and a corresponding target seam segment set, and obtaining a detection task; and controlling each unmanned aerial vehicle to collect laser measurement data of the target seam section in the corresponding flight window based on the detection task so as to execute seam opening size extraction of the corresponding seam section, and updating the seam opening size to the size life cycle curve of the corresponding seam section. According to the scheme of the invention, the problems of insufficient joint section size data acquisition precision, cross-time scale information splitting and the like are effectively solved, and the integrity and dynamics of bridge joint section size evolution monitoring are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of bridge structure detection, and in particular to a method for bridge detection using a swarm of drones and a system for bridge detection using a swarm of drones. Background Art

[0002] As core infrastructure in modern transportation networks, the operational safety of bridges is directly linked to the lifeline of the national economy and the safety of public life and property. As bridges age, problems such as structural aging, stress accumulation, and load fatigue gradually emerge. Joints within bridge structures (such as expansion joints, joints, and structural joints) often become concentrated areas of structural damage. Due to long-term exposure to the combined effects of environmental loads, temperature and humidity fluctuations, and traffic impact, these joints are prone to abnormal opening, delayed closure, crack propagation, or filler delamination. In severe cases, these problems can even affect the structural integrity and performance of the bridge.

[0003] Traditional bridge inspection technologies rely heavily on manual inspections, static monitoring point deployment, or periodic instrument measurements. These technologies suffer from limitations such as low monitoring frequency, limited detection distribution, and delayed data updates. In particular, they struggle to capture the typical characteristics of cracks, such as their dispersed location, complex scale variations, and significant differences in evolution rates. Therefore, establishing a repeatable and quantifiable physical dimension measurement mechanism for bridge cracks has become a critical issue in bridge operation monitoring.

[0004] Among all the indicators related to the state of a joint, the physical dimensions of the joint (especially the width) are key indicators that directly characterize structural stress release, deformation behavior, and disease evolution trends. They possess extremely high structural sensitivity and monitoring value. Accurate extraction of joint dimensions can not only quantify the relative displacement between bridge components but also provide a data basis for structural safety early warning. However, field measurement of joint dimensions faces numerous challenges in practice, such as complex joint morphology, severe reflective disturbances, irregular distribution of measurement areas, and the risk of high-altitude or concealed operations in some areas. These challenges significantly limit the high-precision, high-frequency acquisition of physical dimensions.

[0005] Therefore, there is an urgent need for a technical means to perform high-resolution, spatially continuous, and temporally multi-scale measurements of the physical dimensions of bridge joints, so as to achieve dynamic perception and temporal management of joint dimensions and serve the operation evaluation and risk control of bridge structures. Summary of the Invention

[0006] The purpose of the embodiments of the present invention is to provide a method and system for bridge inspection using a swarm of drones, so as to at least solve the problem that the physical dimensions of bridge joints are difficult to collect and uniformly manage with high precision and continuously at multiple time scales.

[0007] To achieve the above objectives, the present invention provides, in a first aspect, a method for bridge inspection using a swarm of drones, the method comprising: collecting initialization data collected by the swarm of drones, and performing seam type calibration and spatial position partitioning of the target bridge based on the initialization data to form a seam matrix for generating measurement tasks; defining the measurement time window and parameter configuration of each seam in the seam matrix based on each target time scale to generate a multi-time scale measurement task set; performing drone allocation according to the multi-time scale measurement task set to determine the flight window and corresponding target seam set of each drone at each time scale to obtain an inspection task; and controlling each drone to collect laser measurement data of the target seam within the corresponding flight window based on the inspection task to perform seam size extraction of the corresponding seam, and updating the seam size to the size life cycle curve of the corresponding seam.

[0008] Optionally, the initialization data of the drone group is collected, and the joint type calibration and spatial position partitioning of the target bridge are performed based on the initialization data to form a joint matrix for measurement task generation, including: controlling each drone to execute bridge deck image sequence, drone position information and joint edge point cloud data collection during cruising within the target bridge range; performing joint type classification based on the image feature matching results of the joint edge point cloud data and the bridge deck image sequence; wherein the joint type is any one of the following: bridge deck expansion joint, guardrail expansion joint, pier expansion joint, pedestrian expansion joint, bridge head slab gap, hanging basket track structural joint and inclined cable anchor plate joint; performing spatial segment division of each joint based on the position information of each drone; outputting a joint type identifier and a spatial position index based on the joint type and spatial segment of each joint segment, so as to construct a corresponding joint matrix based on the joint type identifier and the spatial position index.

[0009] Optionally, based on each target time scale, the measurement time window and parameter configuration of each seam segment in the seam segment matrix are defined to generate a multi-time scale measurement task set, including: obtaining the time control parameters of the daily variation scale, social cycle scale and seasonal cycle scale from the preset time scale configuration set; setting the measurement start and end time, sampling frequency and response delay threshold at the corresponding time scale as acquisition control parameters based on the time control parameters of each time scale; mapping the acquisition control parameters to the seam segment number of each seam segment in the seam segment matrix one by one, and constructing a multi-time scale measurement task set based on each mapping combination.

[0010] Optionally, each task in the multi-time-scale measurement task set includes: seam segment number, seam segment type, time-scale label, task triggering condition, image acquisition resolution requirement and repetition tolerance; wherein, the time-scale label is used to calibrate the task execution priority; the task triggering condition is used to define the automatic scheduling strategy; the image acquisition resolution requirement is used to guide the UAV configuration to adapt the flight altitude and camera focal length.

[0011] Optionally, drone allocation is performed according to a multi-time-scale measurement task set, including: determining corresponding scheduling parameters for each task in the multi-time-scale measurement task set; calculating the reachability score and task load balance score of each drone within the time window corresponding to the task based on the remaining power, flight capability, historical task load status and image resolution adaptation capability of each drone; on the premise that the task triggering conditions are met, the scheduling score calculation is performed based on the comprehensive reachability score, task load balance score and task priority corresponding to the time scale label to determine the drone with the highest scheduling score for each task, and establish a binding relationship between the corresponding drone and the corresponding task; wherein, each task is bound to only one drone.

[0012] Optionally, the flight window and corresponding target seam segment set of each UAV at each time scale are determined to obtain the detection task, including: counting all task items bound to each UAV based on the binding relationship between each task and the UAV formed with the highest scheduling score; extracting the seam segment number from each task item and summarizing them into the target seam segment set of the UAV; calculating the flight start time, end time and path planning result of the UAV at each time scale according to the measurement time window of the bound task, combined with the flight altitude and heading stabilization time required by the image acquisition resolution requirement, to form a flight window; pairing the target seam segment set with the flight window and storing them as the detection task of the UAV.

[0013] Optionally, based on the detection task, each UAV is controlled to collect laser measurement data of the target seam segment within the corresponding flight window to perform seam size extraction of the corresponding seam segment, including: controlling the UAV to hover stably at the seam segment position within the flight window, and taking an image of the seam area at a preset depression angle; performing edge recognition based on a seam segment boundary line extraction algorithm in the seam area image, and performing target seam segment positioning; according to the target seam segment positioning result, performing laser measurement on the target seam segment to obtain seam size information of the target seam segment.

[0014] Optionally, the size life cycle curve of the seam segment is indexed by the seam segment number, and records the time variation sequence of the seam opening size measured at all time scales for the seam segment; wherein the curve nodes of the size life cycle curve are composed of a timestamp and a corresponding size value.

[0015] Optionally, the seam size is updated to the size life cycle curve of the corresponding seam segment, including: time axis standardization of measurement results at different time scales, classifying data with different measurement frequencies and duration characteristics into corresponding size life cycle curves according to the seam segment numbers, and performing cross-scale differential analysis and trend alignment processing based on overlapping segments between time scales to achieve synchronization of monitoring data at each time scale.

[0016] A second aspect of the present invention provides a system for bridge inspection using a swarm of drones, the system comprising: an acquisition unit for acquiring initialization acquisition data of the drone swarm, and performing seam type calibration and spatial position partitioning of the target bridge based on the initialization acquisition data to form a seam matrix for generating measurement tasks; a processing unit for defining the measurement time window and parameter configuration of each seam in the seam matrix based on each target time scale to generate a multi-time scale measurement task set; a task generation unit for executing drone allocation according to the multi-time scale measurement task set, determining the flight window and corresponding target seam set of each drone at each time scale to obtain an inspection task; and a detection unit for controlling each drone to acquire laser measurement data of the target seam within the corresponding flight window based on the inspection task to extract the seam size of the corresponding seam and update the seam size to the size life cycle curve of the corresponding seam.

[0017] Through the above technical solution, the solution of the present invention constructs a seam segment matrix to perform type calibration and spatial division of the bridge seam area, realizes structured modeling of the detection object, and ensures the comprehensiveness and traceability of subsequent monitoring; combines the multi-time scale definition of measurement task window and parameter configuration to support continuous tracking of bridge structures under different time periods such as daily changes and seasonal responses; through the drone allocation mechanism driven by execution capabilities and mission characteristics, efficient matching of flight missions and detection tasks is achieved; finally, the physical dimensions of the seam are extracted by laser measurement and updated to the dimension life cycle curve, which effectively solves the problems of insufficient accuracy of seam segment dimension data collection and information fragmentation across time scales, and improves the integrity and dynamics of bridge seam segment dimension evolution monitoring.

[0018] Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following detailed description, they are used to explain the embodiments of the present invention, but do not constitute a limitation of the embodiments of the present invention. In the accompanying drawings: Figure 1 This is a flowchart of the steps of a method for bridge inspection using a drone swarm provided by one embodiment of the present invention; Figure 2 This is a system structure diagram of a system for bridge inspection using a swarm of drones, provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0020] The following describes the specific embodiments of the present invention in detail with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only used to illustrate and explain the present invention and are not intended to limit the present invention.

[0021] Figure 1 This is a flowchart of the steps of a method for bridge inspection using a drone swarm provided by an embodiment of the present invention. Figure 1 As shown, an embodiment of the present invention provides a method for bridge inspection using a drone swarm, the method comprising: Step S10: Collect the initialization data of the drone group, and perform seam type calibration and spatial position partitioning of the target bridge based on the initialization data to form a seam matrix for measurement task generation.

[0022] Specifically, each UAV is controlled to collect bridge deck image sequences, UAV position information, and seam edge point cloud data while cruising within the target bridge range; seam type classification is performed based on the image feature matching results of the seam edge point cloud data and the bridge deck image sequence; wherein the seam type is any one of the following: bridge deck expansion joints, guardrail expansion joints, pier expansion joints, sidewalk expansion joints, bridgehead slab gaps, hanging basket track structural joints, and cable anchor plate joints; each seam is spatially segmented based on the position information of each UAV; and a seam type identifier and a spatial position index are output based on the seam type and spatial segment of each seam, so as to construct a corresponding seam matrix based on the seam type identifier and the spatial position index.

[0023] In the practical application of bridge structure inspection, in this embodiment of the present invention, before conducting physical dimension measurements of different types of joints, the locations and structural features of all joints within the target bridge area must be comprehensively and standardizedly identified and organized. To this end, a comprehensive and rigorous process for collecting and structured representation of joint information has been established to ensure the targeted and complete nature of subsequent measurement tasks.

[0024] Specifically, the initialization collection task of the drone swarm is first performed, and each drone is controlled to complete the cruise operation according to the preset flight route within the designated monitoring area of ​​the bridge. During the cruise, three types of data are collected simultaneously: 1) Bridge deck image sequence, used to capture the surface visual information and boundary texture features of the seam area.

[0025] 2) The drone’s own position information, including time-synchronized spatial coordinates and flight attitude angles, is used for spatial positioning and angle correction of collected images.

[0026] 3) The seam edge point cloud data is collected by the lidar or structured light module and is used to construct the geometric boundary and three-dimensional spatial distribution of the seam.

[0027] After basic data collection is complete, the image sequence is registered and fused with the edge point cloud data. First, the boundary features of the joint region are identified using an image feature extraction algorithm (such as SIFT, SURF, or ORB). Feature matching is then performed with the edge changes in the point cloud to form a joint image-point cloud feature set. Based on this, joint type classification is performed. This classification process can be performed using a supervised learning model or a rule-matching model constructed using an existing structural sample library. Different joint types are identified based on differences in visual features, edge topology, and typical morphology. Each identified joint is ultimately labeled as one of the following types: bridge deck expansion joint, guardrail expansion joint, pier expansion joint, sidewalk expansion joint, bridgehead slab gap, gondola track structural joint, or cable anchor plate joint.

[0028] To further support task allocation and path planning, after completing seam type identification, a standardized description of the spatial location of seams is required. Using the location information collected by each drone, each identified seam is partitioned and calibrated within the bridge's panoramic space based on its projected position, actual distribution, and relative distance. Typically, the bridge is divided into multiple spatial segments along its axis. The order, length coverage, and adjacency of the seams within each segment are recorded to generate a spatial segment index.

[0029] Finally, the system combines each joint's type identifier with its spatial segment index to output a structured description format consisting of "joint type identifier plus spatial location index" to construct a joint matrix. This joint matrix organizes all detectable joints in a row-column format, abstracting the joint distribution of the bridge structure into a data structure with statistical boundaries and attribute labels. This supports subsequent operations such as task generation, scheduling planning, and data archiving.

[0030] The above-described process not only ensures unified representation of joint type information but also achieves coordinate-level modeling of spatial distribution, laying the foundation for multiple drones to collaboratively carry out high-precision physical dimension measurement tasks in different regions. Technically, this approach effectively addresses the issues of manual reliance on bridge joint identification, inconsistent spatial positioning, and incomplete data structures. It improves the efficiency of monitoring object organization and the adaptability of scheduling algorithms, demonstrating strong engineering feasibility and platform portability.

[0031] Step S20: Based on each target time scale, define the measurement time window and parameter configuration of each seam segment in the seam segment matrix to generate a multi-time scale measurement task set.

[0032] Specifically, the time control parameters of the daily variation scale, social cycle scale and seasonal cycle scale are obtained from the preset time scale configuration set respectively; based on the time control parameters of each time scale, the measurement start and end time, sampling frequency and response delay threshold under the corresponding time scale are set as acquisition control parameters; the acquisition control parameters are mapped and combined with the segment numbers of each seam segment in the seam segment matrix one by one, and a multi-time scale measurement task set is constructed based on each mapping combination.

[0033] Furthermore, each task in the multi-time-scale measurement task set includes: seam segment number, seam segment type, time-scale label, task trigger condition, image acquisition resolution requirement and repetition tolerance; wherein, the time-scale label is used to calibrate the task execution priority; the task trigger condition is used to define the automatic scheduling strategy; the image acquisition resolution requirement is used to guide the UAV configuration to adapt the flight altitude and camera focal length.

[0034] In this embodiment of the present invention, during the structural monitoring of bridge joints, to effectively capture the changing trends of joint dimensions over different time periods, a measurement task set with multi-timescale control capabilities is required to support monitoring targets at varying granularities, from minutes to quarters. The prerequisite for constructing such a task set is to rationally define the measurement time windows and corresponding parameter configurations for each joint type under different temporal contexts, based on the multi-source variation patterns exhibited during bridge service.

[0035] Specifically, three target time scales were first extracted from a pre-set time scale configuration: diurnal, social, and seasonal. The diurnal scale is primarily used to characterize the regular responses of bridge structures to thermal expansion and contraction, traffic load fluctuations, and other factors within a 24-hour period. For example, this includes the dynamic changes in seam width caused by sudden temperature rises and falls, and morning and evening rush hour commuting loads.

[0036] Furthermore, the social cycle scale is used to monitor the structural response of bridges during social activity cycles, such as the periodic disturbance of the seam width caused by differences in traffic patterns between weekdays and holidays, as well as local stress concentration on the bridge deck caused by large-scale events or emergencies in the short term.

[0037] Furthermore, the seasonal cycle scale is mainly used to evaluate the evolution characteristics of cracks caused by factors such as structural deformation, material aging, and concrete creep caused by long-term climate changes.

[0038] For each target time scale, four control indicators are extracted from its time control parameters: measurement start time, measurement end time, sampling frequency, and response delay threshold. These are collectively referred to as acquisition control parameters. The measurement start and end times define the boundaries of the observation interval at that scale; the sampling frequency represents the number of images required to be acquired within that time period, directly affecting the periodic density of the task; and the response delay threshold defines the maximum tolerance for task execution delays, distinguishing between tasks with strong and weak real-time requirements.

[0039] Each set of acquisition control parameters is then mapped to each joint segment number in the joint segment matrix. The numbers recorded in the joint segment matrix represent the basic joint segment units that have been categorized during the bridge joint segment type calibration and spatial location indexing process. These include bridge deck expansion joints, guardrail expansion joints, pier expansion joints, sidewalk expansion joints, bridgehead slab gaps, gondola track structural joints, and cable anchor plate joints. Each joint segment number combined with a set of acquisition control parameters constitutes a monitoring task to be executed.

[0040] Based on the above-mentioned combination of slit segment numbers and control parameters, a multi-timescale measurement task set is constructed. Each task exists as an independently recorded task unit, and its structure includes the following six key fields: 1) Segment number: used to identify the target segment corresponding to the task, ensuring that the measurement task locates the unique spatial target.

[0041] 2) Seam type: Indicates the structural type of the target seam, providing a basis for the subsequent selection of appropriate processing models and task priority strategies.

[0042] 3) Time scale label: Indicates the time scale type to which the task belongs (e.g., “daily scale,” “social cycle scale,” “seasonal scale”), which is used to guide the UAV scheduling system to prioritize tasks.

[0043] 4) Task trigger conditions: used to define automatic scheduling logic, including fixed-cycle triggering, event-driven triggering (such as measurement anomalies in the previous cycle), or mixed triggering strategies, supporting dynamic adaptation of task execution frequency.

[0044] 5) Image acquisition resolution requirement: This requirement is set based on the fracture segment structure size, surface feature complexity, and fracture boundary resolution requirements, indirectly determining the flight altitude, camera focal length, and stabilization time of the drone required for the mission.

[0045] 6) Repeat tolerance: It is used to measure the acceptable execution redundancy of the task, that is, the tolerance threshold for allowing some tasks to be postponed or merged in multiple consecutive rounds of measurement. This parameter plays an important role in flight path compression and multi-task merging.

[0046] Each task record is organized in a structured manner, facilitating the subsequent sorting, filtering, binding, and scheduling of task sets at various time scales. Furthermore, refined task strategy orchestration can be achieved through the combination of multiple conditions within task fields. For example, within the same seam segment, there may be tasks requiring "daily scale, morning peak sampling, and real-time response" and "seasonal scale, end-of-month sampling, and two-day delay tolerance," each with different execution methods and resource matching requirements.

[0047] This process is highly versatile and scalable. On the one hand, through multi-timescale task configuration, the multi-dimensional coverage of bridge structural response patterns is significantly enhanced, allowing the short-term fluctuations, periodic behaviors, and long-term evolution of structural changes to be simultaneously recorded and tracked. On the other hand, the task set is constructed as an independent field, which not only facilitates cross-platform adaptation but also supports flexible use in subsequent processes such as flight path optimization, drone capacity scheduling, and image data standardization.

[0048] This solution effectively addresses the pain points of traditional bridge monitoring schemes, such as single monitoring frequency, coarse period division, and uncontrollable task parameter configuration. This approach is particularly effective for specialized targets like bridge joints, which exhibit strong structural localization, uneven spatial distribution, and diverse response behaviors. By leveraging the dual mapping of time scales and spatial segments, this approach enables highly refined task configuration. Furthermore, the integrity and clear attribution of task parameters significantly improve the efficiency of subsequent intelligent drone dispatching and the temporal consistency of inspection data, laying a solid foundation for the construction of lifecycle curves for bridge joint dimensions.

[0049] Step S30: Execute UAV allocation according to the multi-time-scale measurement task set, determine the flight window of each UAV at each time scale and the corresponding target seam segment set, and obtain the detection task.

[0050] Specifically, drone allocation is performed according to a multi-time-scale measurement task set, including: determining corresponding scheduling parameters for each task in the multi-time-scale measurement task set; calculating the reachability score and task load balance score of each drone within the time window corresponding to the task based on the remaining power, flight capability, historical task load status and image resolution adaptation capability of each drone; on the premise that the task triggering conditions are met, the scheduling score calculation is performed based on the comprehensive reachability score, task load balance score and task priority corresponding to the time scale label, the drone with the highest scheduling score for each task is determined, and a binding relationship between the corresponding drone and the corresponding task is established; wherein, each task is bound to only one drone.

[0051] Further, the flight window of each unmanned aerial vehicle at each time scale and the corresponding target seam section set are determined, and a detection task is obtained, including: based on the highest scheduling score, the binding relationship between each task and the unmanned aerial vehicle is formed, and all task items bound by each unmanned aerial vehicle are counted; the seam section number is extracted from each task item and is summarized as the target seam section set of the unmanned aerial vehicle; according to the measurement time window of the bound task, the flight height required by the image acquisition resolution requirement and the heading stability time, the flight start time, the flight end time and the path planning result of the unmanned aerial vehicle at each time scale are calculated, and the flight window is formed; the target seam section set and the flight window are paired, and the detection task of the unmanned aerial vehicle is stored.

[0052] In the embodiment of the application, in the multi-time scale detection process of the bridge seam section, in order to guarantee the integrity, accuracy and scheduling efficiency of task execution, an intelligent allocation mechanism based on the task set, taking into account the differences in unmanned aerial vehicle capabilities and time requirements, is needed. Especially in the complex background involving multiple unmanned aerial vehicles, multiple seam sections and multiple time dimensions, a single allocation method cannot meet the comprehensive requirements of task refinement, task collaboration and maximum utilization of resources. Therefore, the present embodiment proposes a multi-time scale measurement task set based unmanned aerial vehicle allocation rule, and based on this, the flight window of each unmanned aerial vehicle at each time scale and the corresponding seam section set are determined, and finally a detection task list with high matching degree and high execution efficiency is output.

[0053] Specifically, first, the multi-time scale measurement task set constructed in the early stage is taken as the input basis. Each task in the task set contains fields such as seam section number, seam section type, time scale label, task trigger condition, image acquisition resolution requirement and repetition tolerance. In order to achieve optimal adaptation of the unmanned aerial vehicle and the task, scheduling parameters need to be extracted from each task, including but not limited to the measurement time window corresponding to the task, the target resolution level, the execution priority corresponding to the time scale label, and the trigger condition.

[0054] On the basis of clear scheduling parameters, the execution capability information of each unmanned aerial vehicle participating in scheduling needs to be obtained. The execution capability information at least includes the following four types: 1) Remaining power: the remaining power of the current unmanned aerial vehicle and the endurance time calculated by the standard flight power, which directly affects whether it has execution capability within the target time window.

[0055] 2) Flight capability: including maximum flight speed, maximum range, height adjustment range and wind resistance level, etc., which is used to determine whether it can meet the flight height and path coverage requirements specified by the task.

[0056] 3) Historical mission load: This refers to the number, distribution density, and execution frequency of the missions that the UAV has undertaken during the current mission cycle, which is used for balanced mission scheduling.

[0057] 4) Image resolution adaptability: This refers to whether the actual imaging capability of the camera meets the image acquisition resolution requirements of the specified task. This capability must match the "Image acquisition resolution requirements" in the task field.

[0058] Based on the aforementioned execution capability parameters, each drone's accessibility score and task load balance score are calculated within the time window corresponding to each mission. The accessibility score is calculated by comprehensively considering the flight path cost (distance and path complexity) from the current location to the target segment, as well as the degree of alignment between the arrival time and the time window, while ensuring that the remaining battery power is sufficient to complete the full mission range. The task load balance score constructs a task density function based on the drone's current task load and the overall distribution of tasks in the task queue. This prioritizes assigning new tasks to drones with lower current loads, thereby avoiding over-concentration of local resources.

[0059] Furthermore, under the premise of meeting the task triggering conditions, the above-mentioned scoring results and the task priority are incorporated into the scheduling scoring function to perform the scheduling scoring calculation.

[0060] Specifically, the scheduling score = α × accessibility score + β × task load balancing score + γ × priority score, where α, β, and γ are adjustable weighting factors, and the priority score is generated by mapping time scale labels (e.g., low for seasonal scale, medium for daily scale, and high for emergencies or morning and evening rush hours). For each task, the scheduling score is calculated by traversing all participating drones, and the drone with the highest score is selected to be bound to the task.

[0061] It's important to note that to ensure the uniqueness and integrity of task execution, each task is only allowed to be bound to one drone after scheduling. This prevents tasks from being executed multiple times and facilitates unified management of task paths and resources. Once all tasks are initially assigned, a complete inspection task list is generated for each drone based on the established "task-drone" binding relationship.

[0062] Specifically, all assigned tasks are traversed and assigned to corresponding drone IDs based on the binding results, forming a preliminary task list for each drone. The corresponding seam segment numbers for all tasks are extracted from the task list and deduplicated to form the target seam segment set for the drone to perform inspection operations. For each task, the shortest path from the starting point to the target seam segment is calculated based on its measurement time window, image acquisition resolution requirements, and flight altitude requirements. Considering the overlap of time windows for multiple tasks, path compression and path splicing are performed between tasks to obtain a coherent and efficient flight route. For each combined path of consecutive tasks, flight control parameters such as the start time, end time, minimum dwell time, turning radius, and stable shooting interval are calculated to form a complete flight window. The target seam segment set is then paired with its corresponding flight window information, and the task number, corresponding seam segment number, execution time, flight parameters, image acquisition parameters, and other information are recorded to form the official inspection task record for the drone.

[0063] This process ensures that each drone is assigned the appropriate task based on its capabilities, while also enabling continuous flight and multi-segment collaborative inspection at varying timescales. Because task scheduling adheres to the principle of "scoring optimization and unique binding," optimal task distribution can be achieved even in scenarios with a limited number of drones.

[0064] Step S40: Based on the detection task, each UAV is controlled to collect laser measurement data of the target seam segment within the corresponding flight window to perform seam size extraction of the corresponding seam segment and update the seam size to the size life cycle curve of the corresponding seam segment.

[0065] Specifically, based on the detection task, each UAV is controlled to collect laser measurement data of the target seam segment within the corresponding flight window to perform seam size extraction of the corresponding seam segment, including: controlling the UAV to stably hover at the seam segment position within the flight window, and taking an image of the seam area at a preset depression angle; performing edge recognition based on a seam segment boundary line extraction algorithm in the seam area image, and performing target seam segment positioning; according to the target seam segment positioning result, performing laser measurement on the target seam segment to obtain the seam size information of the target seam segment.

[0066] Furthermore, the size life cycle curve of the seam segment is indexed by the seam segment number, and records the time-varying sequence of the seam opening size measured at all time scales for the seam segment; wherein, the curve nodes of the size life cycle curve are composed of a timestamp and a corresponding size value.

[0067] Furthermore, the seam size is updated to the size life cycle curve of the corresponding seam segment, including: time axis standardization of the measurement results at different time scales, classifying the data with different measurement frequencies and duration characteristics into the corresponding size life cycle curve according to the seam segment number, and performing cross-scale differential analysis and trend alignment processing based on the overlapping segments between time scales to achieve synchronization of monitoring data at each time scale.

[0068] In an embodiment of the present invention, to achieve long-term, stable, and repeatable monitoring of the physical dimensions of a joint in a bridge structure, each UAV is controlled based on a generated inspection task list, so that it collects laser measurement data of the target joint within the flight window corresponding to the task. Based on image processing and measurement extraction algorithms, the joint size information is obtained. The obtained joint size results are further incorporated into the size life cycle curve of the joint in a time series manner to achieve continuous recording of the structural evolution trend.

[0069] Specifically, based on the flight windows and target seam segments included in the inspection mission, each drone is controlled to enter each flight window in turn, executing the takeoff, hovering, shooting, and return actions set in the mission path planning, and completing image acquisition and size measurement at the target seam segment. During the flight window control phase, priority is given to ensuring the drone's stable hovering capability, especially in wind-load disturbances, narrow areas of bridge structures, or high-low cross-section scenarios. Attitude control is used to keep the drone stable within the preset altitude range to ensure consistency in image acquisition angles and laser ranging paths.

[0070] During image acquisition, the seam area is imaged using a downward-angle vertical camera. This camera can be set to a downward angle of 40° to 70° relative to the horizontal plane, ensuring that the laser ranging path is perpendicular to the seam edge while providing a sufficient field of view to cover the structural features surrounding the seam. The captured image must meet the minimum resolution specified in the image acquisition resolution requirements, for example, a resolution of less than 1mm / pixel, to ensure accurate segmentation of the seam edge by the subsequent boundary extraction algorithm.

[0071] Furthermore, seam segment location and boundary identification are performed based on the captured image. This step first narrows the boundary search range based on the preset spatial index range of the seam segment number. Initial edge detection is performed using the edge gradient characteristics, texture variability, and grayscale distribution in the image. The edge extraction algorithm preferably uses Canny edge detection and edge fusion strategies, and combines the image's surface color characteristics (e.g., concrete color difference, metal plate seam edge) to perform region segmentation and obtain seam segmentation boundaries.

[0072] Once positioning is complete, laser measurement of the target seam is performed using the image boundary as the measurement reference. The laser ranging module performs multi-point distance measurements based on selected measurement points along the target seam boundary. The measurement path is mapped to the image calibration path to obtain physical seam width data in multiple directions. To improve accuracy, a bidirectional laser cross-ranging path is preferably configured. Image pixel size calibration is used to convert pixel values ​​to actual distances. The final output is the seam size value of the target seam at the current moment, including but not limited to average width, maximum opening value, and edge irregularity.

[0073] The acquired seam dimension values, as the result of the current measurement cycle, must be uniformly incorporated into the dimension lifecycle curve for updating. Each dimension lifecycle curve, uniquely indexed by the seam segment number, records the dimension measurements of the corresponding seam segment at all observation times since system deployment. Its core structure is a collection of curve nodes consisting of a timestamp and the corresponding dimension value. Newly added measurement data must undergo consistency preprocessing, including timestamp standardization, deduplication, and outlier removal.

[0074] Given the significant differences in data collection frequency across different timescales, further normalization of the time axis is necessary. For example, data may be collected in hourly increments at the daily scale, while seasonal data may be collected in weekly or monthly increments. To ensure continuity and consistency of comparisons, all measurement nodes are mapped using a unified time axis unit (e.g., day). Periods not covered by measurements are filled in using offset values ​​from adjacent timescales or interpolation.

[0075] Furthermore, to ensure compatibility between data at different time scales, cross-scale differential analysis and trend alignment are required. This process involves calculating the rate of change of fracture dimensions at different time scales, identifying the periodic characteristics of these changes, matching the magnitude and direction of these changes based on overlapping time segments, and performing trend line fitting and correction on measurements with significant deviations. This results in a unified data series that reflects the changing patterns of fracture segments across multiple time dimensions.

[0076] The crack segment dimension lifecycle curve constructed by the present invention not only reflects the structural state of the bridge at a specific point in time but, more importantly, records its entire evolution over time in a continuous and comparable manner. This method significantly improves the accuracy and dimensionality of structural monitoring, avoiding the blind spots and misjudgments caused by traditional periodic manual sampling. Furthermore, the data obtained through the combination of optical imaging and laser measurement has enhanced verification and measurement robustness, providing high-quality data support for bridge structural maintenance, assessment, and early warning.

[0077] Furthermore, based on the dimensional life cycle curve, a threshold alarm mechanism, change trend judgment rules and periodic abnormal behavior identification model can be set for a certain joint section, promoting the upgrade of bridge health monitoring from static interpretation to dynamic prediction, thereby fully supporting the implementation of the "observable, traceable and predictable" smart maintenance concept of bridges.

[0078] Example: Using a main road bridge in a city as the target, a drone swarm-based monitoring mission for the physical dimensions of bridge joints was conducted. The goal was to construct a dimensional lifecycle curve for the joints and enable the perception and archiving of the evolution of bridge joints at multiple time scales. This implementation employed laser ranging and optical image fusion to collect joint dimensions, and multiple drones collaborated to achieve balanced temporal and spatial monitoring.

[0079] During the mission initialization phase, multiple deployed drones were controlled to patrol the bridge structure in sequence, with their flight paths covering all locations where structural joints might exist, including the bridge deck, guardrails, piers, sidewalks, bridgehead slabs, gondola tracks, and cable anchor plates. During this patrol, each drone simultaneously collected a series of bridge deck images, position information (using integrated GNSS and IMU navigation), and 3D point cloud data of the structure's edges. Point cloud data was collected using a rotating LiDAR, while optical images were captured from a bird's-eye view using a 50-megapixel full-frame camera, ensuring that the image and point cloud data could be stitched together.

[0080] After fusing point cloud data with image sequences, the system uses an image feature point matching and point cloud boundary extraction algorithm to identify joint boundaries. Joint types are then classified based on boundary morphology and a database of bridge structure templates. Classification labels include deck expansion joints, guardrail expansion joints, pier joints, sidewalk joints, slab joints, gondola track joints, and cable anchor plate joints. Based on the identification results, each joint is assigned a "joint number" and its spatial location on the bridge structure axis is indexed using its location information. This ultimately generates a joint matrix with a key-value structure of "joint number-joint type-spatial location," providing a standard set of task objectives for subsequent scheduling and monitoring.

[0081] Then, from the pre-set time scale configuration set, we extract monitoring requirements for three typical time scales: minute-to-hour "daily variation scales" (e.g., assessing the impact of morning and evening rush hour traffic loads); daily-to-weekly "social activity cycle scales" (e.g., load variations due to holidays, events, and exhibitions); and monthly-to-quarterly "seasonal variation scales" (e.g., thermal expansion and contraction, freeze-thaw cycles, etc.). For each time scale, we define measurement start and end times, sampling frequency (e.g., minute-level: every 30 minutes; seasonal: every 7 days), and response delay thresholds (a tolerance for data redundancy or delayed responses).

[0082] The above timescale parameters are mapped one by one to each segment number in the segment matrix to generate a complete set of multi-timescale measurement tasks. Each task includes the segment number, segment type, timescale label, task trigger condition (such as "capture when the temperature changes by more than 5°C"), image acquisition resolution requirement (for example, higher than 1mm / pixel), and repetition tolerance (the minimum time interval allowed for repeated acquisitions).

[0083] During the task allocation phase, scheduling parameters are extracted and schedulability analyzed for each task in the task set. For each drone, its accessibility score and load balancing score for each task are calculated based on its remaining battery life, maximum range, historical mission execution (such as the density of the previous round of execution), and its laser measurement and camera parameters. The scheduling score for each task is comprehensively calculated based on the timescale label (priority) and whether the task trigger conditions are met. The drone with the highest scheduling score is selected to establish a task binding relationship. Each task is bound to only one drone to avoid task conflicts and resource duplication.

[0084] After completing the binding, all tasks bound to each drone are counted and the corresponding segment numbers are extracted from the task set to form the drone's target segment set. Based on the time window set for each task, combined with the drone's laser device's ranging height requirements (e.g., vertical range of 3-5 meters) and the shooting attitude adjustment time (heading adjustment time ≥ 3 seconds), the takeoff time, route, operation duration for each segment, and return time are planned for the drone to generate a flight window. Finally, the flight window is bound to the segment set to form the drone's inspection task list.

[0085] During the inspection mission execution phase, each drone enters its flight window within a specific timeframe, automatically takes off, and cruises to the designated seam area according to the mission checklist. After achieving a stable hover, the laser measurement equipment is activated. The seam is imaged at a preset angle (typically 60°), and laser ranging is performed simultaneously to obtain multiple physical parameters, including the maximum seam width and average opening. The image data is aligned with the laser measurement data, and then processed to eliminate outliers. The valid data is then recorded as a single measurement result node.

[0086] The measured data is then written into a "dimensional lifecycle curve" indexed by the seam segment number. This curve records the timestamp and corresponding dimensional value of each acquisition. Data collected at different time scales is standardized to a unified time axis (e.g., converted to daily units). Trend alignment and differential correction are performed in time segments where data overlap to ensure consistency across time scales.

[0087] This dimensional lifecycle curve can be used in subsequent high-level analysis processes, such as structural health diagnosis, trend prediction, and alarm threshold setting. It also enables multi-scale, continuous, and structured management of the physical dimensions of bridge joints. The entire process requires no frequent human intervention, making it highly practical and valuable for engineering promotion.

[0088] Figure 2 This is a system structure diagram of a system for bridge inspection using a drone swarm, provided by one embodiment of the present invention. Figure 2 As shown, an embodiment of the present invention provides a system for bridge inspection using a swarm of drones, the system comprising: an acquisition unit for acquiring initialization acquisition data of the drone swarm, and performing seam type calibration and spatial position partitioning of the target bridge based on the initialization acquisition data to form a seam matrix for generating measurement tasks; a processing unit for defining the measurement time window and parameter configuration of each seam in the seam matrix based on each target time scale to generate a multi-time scale measurement task set; a task generation unit for executing drone allocation according to the multi-time scale measurement task set, determining the flight window and corresponding target seam set of each drone at each time scale to obtain an inspection task; and a detection unit for controlling each drone to acquire laser measurement data of the target seam within the corresponding flight window based on the inspection task to perform seam size extraction of the corresponding seam and update the seam size to the size life cycle curve of the corresponding seam.

[0089] Those skilled in the art will appreciate that all or part of the steps in the methods of the aforementioned embodiments can be accomplished by instructing the relevant hardware through a program. The program is stored in a storage medium and includes a number of instructions for causing a microcontroller, chip, or processor to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0090] The above describes in detail the optional embodiments of the present invention in conjunction with the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details in the above embodiments. Within the technical concept of the embodiments of the present invention, a variety of simple modifications can be made to the technical solutions of the embodiments of the present invention, and these simple modifications all fall within the scope of protection of the embodiments of the present invention. It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner unless there is any contradiction. In order to avoid unnecessary repetition, the embodiments of the present invention will no longer describe the various possible combinations separately.

[0091] In addition, the various embodiments of the present invention may be arbitrarily combined, and as long as they do not violate the concept of the embodiments of the present invention, they should also be regarded as the contents disclosed in the embodiments of the present invention.

Claims

1. A method for bridge inspection using a drone swarm, characterized in that: The method comprises: Collect the initialization data of the drone swarm, and calibrate the seam type and spatial position of the target bridge based on the initialization data to form a seam matrix for measurement task generation; Based on each target time scale, the measurement time window and parameter configuration of each seam segment in the seam segment matrix are defined to generate a multi-time scale measurement task set; Execute UAV allocation according to the multi-time-scale measurement task set, determine the flight window of each UAV at each time scale and the corresponding target seam segment set, and obtain the detection task; Based on the detection task, each UAV is controlled to collect laser measurement data of the target seam segment within the corresponding flight window to perform seam size extraction of the corresponding seam segment and update the seam size to the size life cycle curve of the corresponding seam segment.

2. The method for bridge inspection using a drone swarm according to claim 1, characterized in that: Collect the initialization data of the drone swarm, and calibrate the joint type and spatial position of the target bridge based on the initialization data to form a joint matrix for measurement task generation, including: Control each UAV to collect bridge deck image sequences, UAV position information, and seam edge point cloud data while cruising within the target bridge range; The seam type classification is performed based on the image feature matching results of the seam edge point cloud data and the bridge deck image sequence; The joint segment type is any one of the following: bridge deck expansion joint, guardrail expansion joint, pier expansion joint, sidewalk expansion joint, bridge head slab gap, hanging basket track structural joint and inclined cable anchor plate joint; Divide each seam into spatial segments based on the position information of each UAV; A seam segment type identifier and a spatial position index are output based on the seam segment type and the spatial segment of each seam segment, so as to construct a corresponding seam segment matrix based on the seam segment type identifier and the spatial position index.

3. The method for bridge inspection using a drone swarm according to claim 1, characterized in that: Based on each target time scale, define the measurement time window and parameter configuration of each seam segment in the seam segment matrix to generate a multi-time scale measurement task set, including: The time control parameters of the diurnal variation scale, social cycle scale and seasonal cycle scale are obtained from the preset time scale configuration set; Based on the time control parameters of each time scale, the measurement start and end time, sampling frequency and response delay threshold at the corresponding time scale are set as acquisition control parameters; The acquisition control parameters are mapped and combined with the segment numbers of each segment in the segment matrix one by one, and a multi-time-scale measurement task set is constructed based on each mapping combination.

4. The method for bridge inspection using a drone swarm according to claim 3, characterized in that: Each task in the multi-timescale measurement task set includes: Segment number, segment type, time scale label, task trigger condition, image acquisition resolution requirement and repetition tolerance; among them, The time scale label is used to calibrate the task execution priority; The task triggering condition is used to define the automatic scheduling strategy; The image acquisition resolution requirement is used to guide the UAV configuration to adapt the flight altitude and camera focal length.

5. The method for bridge inspection using a drone swarm according to claim 1, characterized in that: Perform drone assignments based on a multi-timescale measurement mission set, including: Determining corresponding scheduling parameters for each task in the multi-time-scale measurement task set; Based on the remaining battery power, flight capability, historical mission load, and image resolution adaptability of each drone, the reachability score and mission load balance score of each drone within the mission's corresponding time window are calculated. Under the premise of meeting the task triggering conditions, the scheduling score calculation is performed based on the comprehensive reachability score, task load balance score and task priority corresponding to the time scale label, the UAV with the highest scheduling score for each task is determined, and the binding relationship between the corresponding UAV and the corresponding task is established; among them, Each mission is bound to only one drone.

6. The method for bridge inspection using a drone swarm according to claim 5, characterized in that: Determine the flight window of each UAV at each time scale and the corresponding target seam segment set to obtain the detection task, including: Based on the binding relationship between each task and drone formed by the highest scheduling score, all task items bound to each drone are counted; Extract the seam segment numbers from each task item and summarize them into the target seam segment set of the UAV; According to the measurement time window of the bound task, combined with the flight altitude and heading stabilization time required by the image acquisition resolution requirements, the flight start time, end time and path planning results of the UAV at each time scale are calculated to form a flight window; The target seam segment set is paired with the flight window and stored as the detection task of the UAV.

7. The method for bridge inspection using a drone swarm according to claim 1, characterized in that: Based on the detection task, each UAV is controlled to collect laser measurement data of the target seam segment within the corresponding flight window to perform seam size extraction of the corresponding seam segment, including: Control the drone to hover stably at the seam position within the flight window and take images of the seam area at a preset depression angle; Perform edge recognition based on the seam segment boundary line extraction algorithm in the seam area image and perform target seam segment positioning; According to the target seam segment positioning result, laser measurement is performed on the target seam segment to obtain the seam opening size information of the target seam segment.

8. The method for bridge inspection using a drone swarm according to claim 1, characterized in that: The life cycle curve of the seam segment is indexed by the seam segment number and records the time series of the seam size measured at all time scales for the seam segment. The curve nodes of the dimension life cycle curve are composed of a timestamp and the corresponding dimension value.

9. The method for bridge inspection using a drone swarm according to claim 1, characterized in that: Update the seam size to the size life cycle curve of the corresponding seam segment, including: The measurement results at different time scales are time-standardized, and the data with different measurement frequencies and duration characteristics are classified into the corresponding size life cycle curves according to the seam segment numbers. Cross-scale differential analysis and trend alignment processing are performed based on the overlapping segments between time scales to achieve synchronization of monitoring data at each time scale.

10. A system for bridge inspection using a swarm of drones, characterized in that: The system comprises: The acquisition unit is used to collect the initialization data of the drone group, and calibrate the seam type and spatial position of the target bridge based on the initialization data to form a seam matrix for measurement task generation; A processing unit, configured to define a measurement time window and parameter configuration for each seam segment in the seam segment matrix based on each target time scale, so as to generate a multi-time scale measurement task set; The task generation unit is used to perform UAV allocation according to the multi-time-scale measurement task set, determine the flight window of each UAV at each time scale and the corresponding target seam segment set, and obtain the detection task; The detection unit is used to control each UAV to collect laser measurement data of the target seam segment within the corresponding flight window based on the detection task, so as to perform seam size extraction of the corresponding seam segment and update the seam size to the size life cycle curve of the corresponding seam segment.

Images